Professional Documents
Culture Documents
“Victor Babeş” University of Medicine and Pharmacy, Timişoara, 2, Eftimie Murgu sq., 300041
Timişoara, Romania; #e-mail: neagu@umft.ro
Abstract. Dental implant stability depends on the quality of bone in the target site. Although
subjective bone quality assessments are still important, objective measurements of bone density by X-
ray imaging is increasingly appreciated in implant planning. Using conventional computed
tomography (CT), an objective bone density scale was established in terms of mean CT numbers of
various bone types, which characterize their ability to attenuate X-ray beams. Cone-beam computed
tomography (CBCT) is preferred for three-dimensional dental imaging because it is cheaper than CT
and exposes the patient to lower doses of X-rays. The results of bone density measurements by
CBCT, however, are less consistent than CT results: they depend on the type of CBCT device and on
image acquisition parameters. Here we analyzed CBCT images of 46 patients, recorded in identical
conditions by the same type of CBCT unit. We computed the CT numbers of cancellous bone from
400 potential implant sites. Moreover, for each site, we recorded the standard deviation of the CT
numbers of constituent voxels, which is a measure of bone heterogeneity. We classified the sites in
eight groups, according to gender and location (anterior and posterior regions of the mandible and the
maxilla). Based on the one-way ANOVA test and on the Kruskal-Wallis test, we found that
significant differences exist between the mean values of CT numbers and the standard deviations of
CT numbers. Our study suggests that, under identical conditions, CBCT is able to detect differences
in bone density and microstructure. The CBCT scale established here for trabecular bone density and
heterogeneity might be useful for pre-operative evaluation of bone quality.
Key words: CT number, radiodensity, Hounsfield unit (HU), dental implant.
INTRODUCTION
classes, ranging from D1 to D5: D1 bone is dense cortical bone, D2 bone is thick
dense-to-porous cortical bone that wraps a coarse trabecular bone, D3 bone is thin
porous cortical bone that wraps a fine trabecular bone, D4 is fine trabecular bone
within the ridge and minimal or no cortical bone on the crest, whereas D5 is
immature, non-mineralized bone [14, 15].
The quality of bone depends on the anatomical location. The most dense bone
type is found in the anterior mandible, being followed by the posterior mandible,
the anterior maxilla, and the posterior maxilla. The anterior mandible is mainly
composed of D2 bone, but also contains D1 bone in about 6% of the population;
the posterior mandible is made of D2 bone and D3 bone, and rarely contains D1
bone and D4 bone; the anterior maxilla is mainly made of D3 bone, but also of D2
bone (25% occurrence) and D4 bone (10% occurrence); the posterior maxilla is
made of D3 bone and D4 bone, and, occasionally, of D2 bone (10% occurrence)
[15].
An important leap in bone quality estimation was made using computed
tomography (CT), by characterizing bone density in terms of the CT number (or
radiodensity), expressed in Hounsfield units (HU) [18]. The CT number describes
the ability of a substance to attenuate an X-ray beam, ranging from –1000 HU for
air to about 3000 HU for enamel [9]. A significant correlation was found between
the observed radiodensities of the implant sites and their subjective bone density
scores [18]. Further studies [12, 20, 23, 24] have established the ranges of HU
values corresponding to each bone density class: D1 bone > 1250 HU, D2 bone
850–1250 HU, D3 bone 350–850 HU, D4 bone 150–350 HU and D5 bone < 150
HU [15].
Nevertheless, for the three-dimensional (3D) visualization of hard tissues
from the oral environment CT is not the imaging technique of choice; it exposes
the patient to relatively high doses of ionizing radiation and requires expensive
equipment. Instead, cone beam computed tomography (CBCT) is preferred because
it involves less radiation exposure (by about one order of magnitude) and requires a
device that is about five times cheaper than a full-body CT scanner [7, 22]. While
in a conventional CT scanner the body part of interest is traversed by a fan-shaped
beam of X-rays that falls on a linear arrangement of detectors, in a CBCT device
the investigated body part is traversed by a cone-shaped beam of X-rays that falls
on a flat panel detector (such as a screen of scintillator crystals placed on a matrix
of photodiodes embedded in solid-state amorphous silicon; the scintillator crystals
convert X-rays into visible light, which is detected by the photodiodes) [13].
CBCT is routinely used to visualize the bony structures of the head and neck
with typical voxel sizes as small as 0.2×0.2×0.2 mm. The image reconstruction
software associates to each voxel a HU value, also called CT number in the
literature, although it is recorded using a CBCT device. Commercially available
3 Trabecular bone assessed using CBCT 229
PATIENTS
For this retrospective study 46 edentulous patients were selected from the
Mall Dental Clinic Timisoara, Romania, who had undergone a CBCT investigation,
between 2011 and 2013, for reasons that had nothing to do with the present study.
No patient was subject to ionizing radiations for the sake of this work.
To enroll patients, we relied on the following inclusion criteria: (i) CBCT
images were recorded using a tube voltage of 84 kV and a current intensity of
14 mA, (ii) the CBCT images were recorded with a cylindrical field of view (FOV)
of 80 mm both in diameter and height (Figure 1), (iii) the voxel size was
0.2×0.2×0.2 mm, (iv) the dental arches were positioned similarly in the FOV, as
depicted in Figure 1.
Coronal Sagittal
Axial
We excluded from the study patients who had pathologic and/or traumatic
lesions of the jaws.
To assure gender balance, an equal number of men and women were enrolled
in the study. Moreover, the number of investigated implant sites was the same for
each gender.
5 Trabecular bone assessed using CBCT 231
Fig. 2. A typical CBCT recording included in this study, taken with the
cylindrical field of view highlighted in Figure 1. The region of interest is a
rectangular slab of trabecular bone located in edentulous regions of the
alveolar ridge. The Romexis software characterizes the selected volume
(white text in frame), listing its size, the mean value of the CT numbers
associated to the voxels that compose it (labeled as “Mediu”), the standard
deviation of the CT numbers of the constituent voxels (labeled as “Deviaţie
std”), and the range of CT numbers of the voxels from the volume (listed in
the format R [smallest CT number, largest CT number]).
232 O.T. David, M. Leretter, A. Neagu 6
STATISTICAL ANALYSIS
RESULTS
Table 1
Descriptive statistics of the CT numbers of the investigated trabecular bone volumes. The last column
lists the P-values of the normality test
Gender Anatomical Number of Mean CT Standard Skewness Kurtosis P
location sites number error
(HU) (HU)
Anterior maxilla 34 354 30.0 0.3235 3.401 > 0.5
Posterior maxilla 66 193 17.4 0.6999 2.828 0.046
Women
Anterior mandible 27 514 30.5 0.4455 2.086 0.185
Posterior mandible 73 234 17.5 0.9581 3.369 0.012
Anterior maxilla 40 473 33.5 –0.1010 3.289 > 0.5
Posterior maxilla 60 250 18.7 0.8194 3.277 0.031
Men
Anterior mandible 28 521 22.0 0.7559 3.806 0.071
Posterior mandible 72 389 28.0 0.6159 2.689 0.057
Table 2
Descriptive statistics of the standard deviations (SD) of CT numbers of voxels that compose the
image of the trabecular bone of interest. The last column lists the P-values of the normality test
Gender Anatomical Number of Mean SD of Standard Skewness Kurtosis P
location sites CT numbers error
(HU) (HU)
Anterior maxilla 34 206 7.57 –0.165 2.544 > 0.5
Posterior maxilla 66 176 5.08 0.4987 2.765 0.133
Women
Anterior mandible 27 243 10.1 0.0904 2.539 > 0.5
Posterior mandible 73 212 5.13 0.2695 2.282 0.184
Anterior maxilla 40 208 5.67 –0.1041 2.159 0.389
Posterior maxilla 60 193 4.66 0.2782 2.224 0.193
Men
Anterior mandible 28 247 11.3 0.07911 2.495 > 0.5
Posterior mandible 72 220 5.44 0.4889 3.284 0.120
box), whereas the whiskers span the interval between extreme data points not
classified as outliers. Individual markers (+ signs) represent outliers. The right
panel represents similar box plots of the standard deviations of CT numbers of the
constituent voxels.
Fig. 3. Box plots of the medians of the CT numbers (left) and the medians of the standard deviations of CT
numbers (right) obtained for eight groups of trabecular bone volumes: 1 – women anterior maxilla,
2 – women posterior maxilla, 3 – women anterior mandible, 4 – women posterior mandible, 5 – men
anterior maxilla, 6 – men posterior maxilla, 7 – men anterior mandible, 8 – men posterior mandible.
Table 3 presents the results of this test as a standard one-way ANOVA table
[5]. In the terminology of ANOVA distinct groups are called treatments. The first
row of Table 3 refers to treatments: 7 is the number of degrees of freedom (equal to
the number of treatments minus 1), the number listed in the next column (SS) is the
treatment sum of squares, it is followed by the treatment mean square (in column
MS), by the F-statistic given by the ratio of the variation among sample means to
the variation within the samples (column F), and the corresponding P-value. The
second row of Table 2 refers to errors: the number of degrees of freedom is the
number of observations minus the number of treatments (column df); it is followed
by the error sum of squares (column SS), and the error mean square (column MS,
given by the ratio of the error sum of squares to the number of degrees of freedom
for errors).
Table 3
The results of the one-way ANOVA test for the mean CT numbers of the voxels that compose the
trabecular bone volumes under study
Source df SS MS F P
Groups 7 5.3284×106 7.6120×105 25.03 3.2168×10–28
Error 392 1.1922×107 3.0413×104 – –
Total 399 1.7250×107 – – –
The large value of F suggests that the discrepancies between sample means
might not stem from discrepancies between the values that belong to individual
9 Trabecular bone assessed using CBCT 235
samples. Can we conclude that there are significant differences between the
means? In a hypothesis test, the decision is based on the P-value, which is the
probability of obtaining the observed results when H0 is true. The small value of P,
(more precisely, P < 0.05 at a 5% significance level) indicates that H0 can be
rejected in favor of Ha; that is there are significant differences between certain
mean values.
One-way ANOVA was also used to test whether there are significant
differences between the mean values of the standard deviations of CT numbers of
the voxels that compose the 3D images of the bone slabs under study. Table 4 lists
the results of this test, indicating that not all means are equal.
Table 4
The results of the one-way ANOVA test for the mean values of the standard deviations of CT
numbers of the voxels that compose the investigated trabecular bone volumes
Source df SS MS F P
Groups 7 1.6811×105 2.4016×104 12.41 2.2605×10–14
Error 392 7.5872×105 1.9355×103 – –
Total 399 9.2683×105 – – –
Table 5
The results of the Kruskal-Wallis test for the mean CT numbers of the voxels from trabecular bone
samples
Source df SS MS Chi-sq P
Groups 7 1.7695×106 2.5279×105 132.39 1.9938×10–25
Error 392 3.5637×106 9.0910×103 – –
Total 399 5.3332×106 – – –
Table 6
The results of the Kruskal-Wallis test for the mean values of the standard deviations of CT numbers
of the voxels from trabecular bone samples
Source df SS MS Chi-sq P
Groups 7 8.6215×105 1.2316×105 64.51 1.8902×10–11
Error 392 4.4706×106 1.1405×104 – –
Total 399 5.3328×106 – – –
Since the ANOVA test is based on the assumption that the data are normally
distributed [5], and the normality test gave a borderline result for groups 2, 4, 6 and
8 (see last column of Table 1, values corresponding to posterior regions of the
dental arches), we also performed a Kruskal-Wallis test for the same null
hypothesis (H0) and alternative hypothesis (Ha). Tables 5 and 6 show the results of
236 O.T. David, M. Leretter, A. Neagu 10
the Kruskal-Wallis test for the mean CT numbers and the mean values of the
standard deviations of CT numbers, respectively.
Hence, both the one-way ANOVA test and the Kruskal-Wallis test indicate
that not all means are equal. These tests, however, do not give any hint on which
pairs of means differ from each other. To address this question, we performed
multiple comparisons using the multcompare function from the Statistics Toolbox
of MATLAB. The results of multiple comparisons are shown on Figure 4.
Fig. 4. Multiple comparisons of the mean values of CT numbers (A) and standard
deviations of CT numbers (B) based on the one-way ANOVA test; multiple
comparisons of the mean ranks of CT numbers (C) and mean ranks of standard
deviations of CT numbers (D) based on the Kruskal-Wallis test. The groups of
trabecular bone regions under study are: 1 – women anterior maxilla, 2 – women
posterior maxilla, 3 – women anterior mandible, 4 – women posterior mandible,
5 – men anterior maxilla, 6 – men posterior maxilla, 7 – men anterior mandible,
8 – men posterior mandible.
DISCUSSION
density was found to be the highest in the anterior mandible (530 HU), followed by
the anterior maxilla (516 HU), the posterior mandible (359 HU) and the posterior
maxilla (332 HU) [8]. In what concerns the anterior mandible our results are closer
to the ones of ref. [8], whereas for the posterior maxilla our results are closer to the
ones of ref. [6].
In their study of trabecular bone from the mandible by CBCT and multislice
helical CT [17], Naitoh et al. found a strong correlation between voxel values (HU
values) measured via CBCT and bone mineral density obtained by helical CT.
Nevertheless, their CBCT-derived voxel values, of 655 HU to 747 HU for the
anterior mandible, and 519 HU for the posterior mandible, are higher than ours,
although their study group consisted mainly of women (4 men and 12 women).
This discrepancy might stem from using a different type of CBCT unit, Alphard
VEGA (Asahi Roentgen Ind. Co., Japan), and different settings [17].
Bone density measurements in HU units were performed on 236 potential
implant sites identified in CBCT scans of 74 men and 54 women, recorded by the
same type of unit as ours, at slightly different settings (90 kV, 14 mA, and voxel
size of 0.2×0.2×0.2 mm) [10]. Although it does not focus on trabecular bone, but
on bone located in the vicinity of a cylindrical implant simulated using the
Simplant software, a comparison with our results is justified for D4 bone, which is
basically cancellous. Moreover, it is reasonable to assume that the simulated
implants, 4.1 mm in diameter and 10 mm in length, placed in edentulous spans of
the posterior jaws are mainly surrounded by trabecular bone. Indeed, the mean CT
numbers of the posterior mandible (394 HU) and the posterior maxilla (220 HU)
are comparable with our results. The slightly higher HU value observed by Hao et
al. in the posterior mandible and the significantly higher HU values observed in the
anterior jaws might stem from cortical bone contributions [10].
Significant differences between means can be identified from Figure 4
keeping in mind that the Kruskal-Wallis test is less efficient than the one-way
ANOVA test when the assumptions for applying the latter hold [5]. Thus, for
analyzing HU values, for which the normality test gave borderline results, the
Kruskal-Wallis test is more reliable. This was the statistical test of choice also in
reference [8] for the analysis of CT numbers of trabecular bone acquired by
conventional CT. In our study, according to the last column of Table 1, HU values
of samples from the anterior portion of the dental arches are normally distributed,
while the HU values of samples from the posterior regions are not (excepting the
posterior mandible of men, whose P-value, however, barely exceeds 0.05).
Therefore, to identify mean HU values that differ significantly from each other, it
is preferable to rely on the confidence intervals of mean ranks shown in Figure 4C,
derived from multiple comparisons based on the Kruskal-Wallis test (instead of
Figure 4A, which stem from the one-way ANOVA test). For instance, looking at
the region delimited by the dotted lines from Figure 4C, we infer that the mean CT
number of group 1 (anterior maxilla of women) differs significantly from the mean
13 Trabecular bone assessed using CBCT 239
CT numbers of groups 2, 4 and 7. The same figure shows that the mean CT number
of trabecular bone from the anterior mandible of men (521 HU) is significantly
higher than the mean CT numbers of other groups, except for the anterior mandible
of women (514 HU) and anterior maxilla of men (473 HU). Moreover, we remark
that, for each anatomical region, the mean CT number is larger for men than for
women, but their difference is not significant from the statistical point of view.
For analyzing standard deviations of HU values (which are normally
distributed according to Table 2) multiple comparisons based on the one-way
ANOVA test (Figure 4B) are more powerful than those based on the Kruskal-
Wallis test (Figure 4D) [5]. From the dotted lines of Figure 4B we infer that,
regarding standard deviations of CT numbers of the constituent voxels, the mean of
group 7 (anterior mandible of men) is significantly larger than the means of other
groups, except for groups 3 (anterior mandible of women) and 8 (posterior
mandible of men). Thus, standard deviations of CT numbers of the constituent
voxels discriminate between coarse and fine trabecular bone. Just as in the case of
mean voxel values, for each anatomical region, the standard deviation of CT
numbers is larger for men than for women, but not significantly so from the
statistical point of view.
CONCLUSIONS
Acknowledgements. This paper was published under the frame of the European Social Fund,
Human Resources Development Operational Programme 2007–2013, project No.
POSDRU/159/1.5/136893.
REFERENCES